Unless you've been living under a rock, you would've noticed that artificial neural networks are now used everywhere. They're impacting our everyday lives, from performing predictive tasks such as recommendations, facial recognition and object classification, to generative tasks such as machine translation and image, sound, video generation. But with all of these advances, the impressive feats in deep learning required a substantial amount of sophisticated engineering effort. Even if we look at the early AlexNet from 2012, which made deep learning famous when it won the ImageNet competition back then, we can see the careful engineering decisions that were involved in its design. Modern networks are often even more sophisticated, and require a pipeline that spans network architecture and careful training schemes.

The discovery of complex multicellular organism development took millions of years of evolution. The genome of such a multicellular organism guides the development of its body from a single cell, including its control system. Our goal is to imitate this natural process using a single neural cellular automaton (NCA) as a genome for modular robotic agents. In the introduced approach, called Neural Cellular Robot Substrate (NCRS), a single NCA guides the growth of a robot and the cellular activity which controls the robot during deployment. We also introduce three benchmark environments, which test the ability of the approach to grow different robot morphologies. In this paper, NCRSs are trained with covariance matrix adaptation evolution strategy (CMA-ES), and covariance matrix adaptation MAP-Elites (CMA-ME) for quality diversity, which we show leads to more diverse robot morphologies with higher fitness scores. While the NCRS can solve the easier tasks from our benchmark environments, the success rate reduces when the difficulty of the task increases. We discuss directions for future work that may facilitate the use of the NCRS approach for more complex domains.

Reaction-Diffusion (RD) systems provide a computational framework that governs many pattern formation processes in nature. Current RD system design practices boil down to trial-and-error parameter search. We propose a differentiable optimization method for learning the RD system parameters to perform example-based texture synthesis on a 2D plane. We do this by representing the RD system as a variant of Neural Cellular Automata and using task-specific differentiable loss functions. RD systems generated by our method exhibit robust, non-trivial 'life-like' behavior.

Neural cellular automata (CA) is a kind of cellular automaton (Figure 1). While cellular automata determine the state transition rule of cells by hand-making, neural CA obtains the transition rule by training a neural network. Recently, this neural CA has been shown to be a powerful tool in morphogenesis [1]. Mordvintsev et al. trained a neural CA to grow complex two-dimensional images starting from a few initial cells. Furthermore, authors also successfully trained it to regenerate a target pattern even if part of it is removed.